The IoT market is increasing consumer touchpoints as one axis of edge AI. Accordingly, Arm has unveiled a new portfolio for next-generation edge AI that boasts edge AI technology leadership in the IoT market.

On the 27th, Arm held a press conference at The Plaza Hotel in Jung-gu, Seoul and announced the Armv9 Edge AI platform equipped with the Arm Cortex-A320 CPU and the Arm Ethos-U85 NPU, an accelerator for edge AI.

The Armv9-based A320 CPU, which replaces the previous generation A35 and A53, has been explained as being strengthened compared to the previous generation in terms of changing and evolving edge AI workloads and deployment flexibility. Targeting the IoT market, these high-efficiency edge AI processors have advantages such as reduced power consumption and development flexibility for time to market (TTM).

Compared to the A35, single-core machine learning has increased by 10 times, scalar performance has increased by 30%, and power efficiency has improved by 50% compared to the A520. He added that it is possible to run AI models with 1 billion parameters on edge devices.

“Arm believes that there are limits to processing all AI in the cloud, and as the AI revolution progresses rapidly, IoT workloads are becoming increasingly complex,” said Hwang Sun-wook, president of Arm Korea. “Developers are being asked to develop safe and flexible AI that can run on devices with more computing performance and less power consumption.”

Accordingly, Arm expects to provide customers with the benefits of expanding the ease of edge AI application development and shortening the IoT development period through the next-generation Armv9 edge AI platform and A320 processor at the center of the edge AI ecosystem. The A320-based silicon chip is expected to be released around 2026.

FAE Director Seong-Hoon Jeong said, “When looking at machine learning utilization cases, the value of data within devices and the software that processes it are increasing as they become more complex,” and explained that security is being emphasized more accordingly.

The A320 is designed to leverage security and AI performance for IoT based on Armv9, and in particular, the security of the Armv9.2 architecture includes advanced security features such as △PAC (Pointer Authentication), △BTI (Branch Target Identification), and △MTE (Memory Tagging Extension) in an ultra-small Cortex-A device.He added that it also provides:

Because edge devices operate in exposed environments and process sensitive data, security performance is important to prevent data leakage at the edge.

Arm also addressed the development and deployment complexity that is a barrier to the adoption of edge AI with Arm Kleidi. Arm Kleidi, a set of compute libraries for AI framework developers, has been expanded to IoT, strengthening the software ecosystem.

“For existing customers looking to further enhance their previous-generation products, we believe the next-generation Arm Edge AI platform will be an efficient edge device,” said Hwang Sun-wook, president. “For example, a doorbell will not only provide a simple bell notification, but will also provide specific information such as notifying the arrival of a delivery item and detecting motion.”